Tracking systems are configured to track a moving object in a sequence of images or video. One type of tracking system is a feature-based tracking system. The feature-based tracking system extracts feature data (e.g., data of “point features”) from a region in an image that includes object. The feature data includes data representing edges, corners, textures, size, or other aspects of the object that distinguish the object from background pixels. The extracted feature data is representative of the object, and the feature-based tracking system searches subsequent images of the sequence of images for regions that have similar feature data in order to track the object. When the sequence of images has a high quality (e.g., a high number of pixels) and when the object has sufficient features (e.g., texture, size, corners, etc.) to distinguish the object from the background pixels, the feature-based tracker may track the object successfully in the subsequent images. However, when the sequence of images has a low quality (e.g., a small number of pixels) or when the object lacks sufficient features to distinguish the object from the background pixels, the feature-based tracking system may be unable to track the object in the subsequent images.
Another type of tracking system is an instance-based tracking system, also referred to as a multiple instance learning (MIL) tracking system. The instance-based tracking system determines instances (e.g., small “snapshots”) of regions in the image that include the object (e.g., positive instances) and instance of regions surrounding the object (e.g., negative instances). The instance-based tracking uses the instances to track the object in subsequent images of the sequence of images (e.g., by identifying similar positive instances or negative instances). Because the instances may indicate a difference between the object and surrounding regions when the object has relatively few features or when the sequence of images has a low quality (e.g., a small number of pixels), the instance-based tracking system may be able to track the object in subsequent images that are problematic for the feature-based tracking system. However, when an image includes multiple regions that appear similar to the object, the instance-based tracking system may be unable to distinguish the object from a region having a similar appearance. Additionally, when one or more characteristics of the object changes in the subsequent images, the instances may not be similar enough to the appearance of the object in the subsequent images, causing the instance-based tracking system to lose track of the object.